[Regression] "die;" does not always propagate exception #11996
Comments
From anthony@derobert.netCreated by anthony@derobert.netThis is a bug report for perl from anthony@derobert.net, ----------------------------------------------------------------- NOTE: This is also Debian bug 663158 "die" does not always propagate the exception in $@. The below test foo at survey.cgi line 22 instead it prints Died at survey.cgi line 19. This worked in 5.12.x and 5.10.x. I have tested this on Debian's compiled perl (details below from Performing a git bisect, I found the regression was introduced by commit c5df309 SV-based interfaces for dieing and warning Test program. Note that it's very fragile. Take off -T, get rid of the ===== BEGIN TEST PROGRAM ===== use Carp qw(confess); my $CGI; sub send_template; my $dat = <<DATA; eval { run_template($CGI->param('survey_no')); }; sub run_template { Note that you can find the full bisect log, the test programs used for Perl Info |
From @cpansproutOn Fri Mar 09 09:23:57 2012, derobert wrote:
That’s a pretty nasty regression. Can we make this a blocker for 5.16? -- Father Chrysostomos |
|
The RT System itself - Status changed from 'new' to 'open' |
From @tonycozOn Fri, Mar 09, 2012 at 01:09:54PM -0800, Father Chrysostomos via RT wrote:
I think it should be a blocker too. Tony |
From @tonycozOn Sat, Mar 10, 2012 at 03:21:29PM +1100, Tony Cook wrote:
A simpler test case: #!/usr/bin/perl -Tw |
From @tonycozOn Sun, Mar 11, 2012 at 01:19:36PM +1100, Tony Cook wrote:
Attached: regression test and possible fix. It may be that the code should be checking for magic earlier, but with Tony |
From @tonycoz0001-rt-111654-TODO-test-for-tainted-die-propagation.patchFrom 6c3816fb002c13b546764174e8576b2012725cdf Mon Sep 17 00:00:00 2001
From: Tony Cook <tony@develop-help.com>
Date: Sun, 11 Mar 2012 14:27:29 +1100
Subject: [PATCH 1/2] [rt #111654] TODO test for tainted die propagation
---
t/op/taint.t | 11 ++++++++++-
1 files changed, 10 insertions(+), 1 deletions(-)
diff --git a/t/op/taint.t b/t/op/taint.t
index 1b75439..095c991 100644
--- a/t/op/taint.t
+++ b/t/op/taint.t
@@ -17,7 +17,7 @@ BEGIN {
use strict;
use Config;
-plan tests => 793;
+plan tests => 794;
$| = 1;
@@ -2197,6 +2197,15 @@ pass("no death when TARG of ref is tainted");
is_tainted "\F$utf8", "under locale, \\Futf8 taints the result";
}
+{ # 111654
+ local $::TODO = "RT #111654";
+ eval {
+ eval { die "Test\n".substr($ENV{PATH}, 0, 0); };
+ die;
+ };
+ like($@, qr/^Test\n\t\.\.\.propagated at /, "error should be propagated");
+}
+
# This may bomb out with the alarm signal so keep it last
SKIP: {
skip "No alarm()" unless $Config{d_alarm};
--
1.7.2.5
|
From @tonycoz0002-rt-111654-properly-propgate-tainted-errors.patchFrom e00ad30ae2b8f545e4f96cab8e76ccd74840e687 Mon Sep 17 00:00:00 2001
From: Tony Cook <tony@develop-help.com>
Date: Sun, 11 Mar 2012 14:38:57 +1100
Subject: [PATCH 2/2] [rt #111654] properly propgate tainted errors
A magic value (such as a tainted string) may not have POK set, so call
SvPV() to find out if there's something in ERRSV to report.
Possibly this should be using SvPV_nomg(), but this is the first
request for magic in this code. Maybe the code above should be
calling SvGETMAGIC() before checking SvROK().
---
pp_sys.c | 2 +-
t/op/taint.t | 1 -
2 files changed, 1 insertions(+), 2 deletions(-)
diff --git a/pp_sys.c b/pp_sys.c
index 63fbd05..83fe34d 100644
--- a/pp_sys.c
+++ b/pp_sys.c
@@ -500,7 +500,7 @@ PP(pp_die)
}
}
}
- else if (SvPOK(ERRSV) && SvCUR(ERRSV)) {
+ else if (SvPV(ERRSV, len) && len) {
exsv = sv_mortalcopy(ERRSV);
sv_catpvs(exsv, "\t...propagated");
}
diff --git a/t/op/taint.t b/t/op/taint.t
index 095c991..9cea740 100644
--- a/t/op/taint.t
+++ b/t/op/taint.t
@@ -2198,7 +2198,6 @@ pass("no death when TARG of ref is tainted");
}
{ # 111654
- local $::TODO = "RT #111654";
eval {
eval { die "Test\n".substr($ENV{PATH}, 0, 0); };
die;
--
1.7.2.5
|
From @rjbsThis blocks 5.16.0 |
From @cpansproutOn Sat Mar 10 19:45:35 2012, tonyc wrote:
Checking for magic earlier before the SvROK check would be the correct If you are going to use SvPV, I believe it always returns true, so what Interestingly, using SvPV fixes this bug, too: $ perl5.14.0 -e '$@ = 3; die' -- Father Chrysostomos |
From zefram@fysh.orgTony Cook wrote:
Yes, this looks right. There are many kinds of scalar that are not POK -zefram |
|
@tonycoz - Status changed from 'open' to 'resolved' |
From @jmdhOn Sun, Mar 11, 2012 at 04:39:27PM -0700, Tony Cook via RT wrote:
And queued in Debian's git (simple backport to 5.14 which just needed http://anonscm.debian.org/gitweb/?p=perl/perl.git;a=commit;h=e6d732e0cdea0d6906b732628f6a55f93f91f633 -- |
Migrated from rt.perl.org#111654 (status was 'resolved')
Searchable as RT111654$
The text was updated successfully, but these errors were encountered: